Comment by SpicyLemonZest

9 hours ago

Frontier model developers try to check for memorization. But until AI interpretability is a fully solved problem, how can you really know whether it actually didn't memorize or your memorization check wasn't right?

0 comments